Very Peculiar and Erroneous Auto-Tune Behavior

ArduCopter Copter 4.3
1

I’ve been getting some very peculiar and erroneous behaviors with Auto-Tune on Copter 4.3.

Some times one of PIDs axis are inconsistent.

Some times a couple of parameters for an axis don’t get stored. (observed by comparing the MavLink message report or the Auto-Tune PIDs with what actually got stored in the parameters)

Some times the twitching is a little bizarre - and autotune has to be aborted before all axis are complete.

I don’t have detailed enough notes to totally reconstruct my tests - but I have been keeping records of PIDs from successive tunes.

pids that make sense
pids that make sense1056×771 51 KB

The shaded values are the ones that make the most sense - and they one’s I’ll try again with tomorrow.

I flew a test mission after completing the autotune in COLUMN-T - where the roll PIDs don’t make sense. The MavExplorer attitude plot confirms that roll is wrong.

roll is wrong
roll is wrong1920×1080 315 KB

Until it’s sorted out - I urge caution.

2 Likes
2

Did you graph your vibes ? Usually vibrations ruin autotune runs. Happened to me last week.

1 Like
3

No - but I can easily do it now.

Here’s the vibe plot from an AutoTune were the pitch and roll twitches were OK, but then the yaw twitches became erratic and I hand to cancel the AutoTune sessions.

autotune - bad yaw twitches
autotune - bad yaw twitches1920×1080 291 KB

And here’s the vibe plot from when roll twitches were OK, but I had to cancel autotune when the pitch twitches became erratic.

autotune - bad pitch twitches
autotune - bad pitch twitches1920×1080 263 KB

The BIN files for these two flights are on dropbox: